Yes, poker training works, and you do not have to take our word for it. We pulled every synced training history in the GTO Gecko database, 673,118 solver-graded decisions from real players, and followed fixed groups of players from their first decision onward. Players make 27% fewer mistakes by their 25th session and cut their EV loss per decision by 28% inside 1,000 trained hands. Among long-haul players, 81% of the total measured improvement arrives inside the first 750 decisions. This post shows the numbers, the charts, and the published research that explains why drilling works so well.
Disclosure: GTO Gecko makes the trainer this data comes from. Everything below is observational product data, described honestly, with the methodology and its limits stated in plain language. None of it is a promise about your win rate at the tables.
How Did We Measure Improvement?
Every decision a player makes in the GTO Gecko trainer is graded against the solver's strategy for that exact spot and stored with its EV cost in big blinds. That gives us something rare in poker: a clean, decision-level record of how real players change as they practice. The dataset behind this post: 673,118 graded decisions from 317 players with synced training histories, computed on August 8, 2026.
Three rules kept the analysis honest. First, improvement curves use fixed cohorts: the same players in every data point, so trends are never an artifact of who joined or quit. Second, a mistake means the grader scored the decision wrong or blunder (a low-frequency action losing more than 1% of the pot; blunders lose more than 5%). Third, one bias runs against us: the trainer adapts and serves harder spots as you improve, so if anything these numbers understate real progress. If you want to know how the underlying solutions are produced, that is covered in how our solutions are made.
An honest data post should also point at the counter-evidence. The best-known study concluding that poker is a game of chance (Meyer and colleagues, 2013) had each participant play exactly 60 hands, deep inside the window where the 456-million-hand study below agrees that luck rules. And spacing researchers find smaller scheduling benefits on complex tasks than on simple ones, so we treat the practice-scheduling literature as directional support, not gospel. The improvement numbers above stand on their own data either way.
Do Players Actually Make Fewer Mistakes?
They do, and the effect is big. We followed the 100 players who completed at least 25 training sessions. In their first session they misplayed 14.1% of decisions. By session 25 that was down to 10.3%, a 27% drop in mistake rate, and the improvement is statistically solid (99.7% bootstrap confidence). The median session took under 5 minutes.
How Much EV Do Players Stop Losing?
About 28% less per decision, within the first 1,000 trained hands. Tracking the 99 players who reached 1,000 trained decisions, average EV lost per decision fell from 0.268 BB in their first 100 decisions to 0.194 BB by decisions 901 to 1,000.
Look closer and the improvement has a shape: training deletes the disasters first. The EV bled in big mistakes (spots costing 2 BB or more) dropped 38% across those 1,000 decisions, while small imprecisions changed little. That matches how the money actually moves: across our whole database, the average blunder torched 2.75 big blinds. A strong winning player earns about 5 BB per 100 hands, so a single unnoticed blunder erases half of that. Killing one recurring blunder is worth more than polishing fifty small spots.
How Big Can the Improvement Get?
The averages hide the upside. Comparing each player's first 250 decisions with their latest 250, 43% of players cut their EV loss by 30% or more, and roughly 1 in 3 cut it in half. The top tenth of improvers cut it by roughly 84%.
Does Training a Little Every Day Beat Cramming?
By a factor of six. Players who trained on 10 or more days in their first month gained a median of 3.0 accuracy points. Players who trained on 4 days or fewer gained 0.5. Frequency beat intensity, exactly as the learning research predicts: in the largest review of practice scheduling ever conducted (839 assessments), spacing beat cramming in 259 of 271 direct comparisons. The classic typing-training experiment makes it concrete: trainees practicing one hour a day reached the target skill in 30% fewer total hours than the four-hours-a-day group, on identical total practice, and typed 22% faster at the end.
Why Is a Trainer Faster Than Just Playing?
Because almost none of your time at the table is decision time. A live table deals roughly 25 to 30 hands per hour, and you fold most of them. An online table deals 75 to 100. The median GTO Gecko player makes 282 graded decisions per hour of active training, each with instant feedback: the correct action, and what your choice cost in EV. One focused hour buys you a week's worth of live-table decisions.
Instant feedback is what makes that hour count, but only the right kind. The largest meta-analysis of feedback ever conducted (607 effect sizes, 23,663 observations) found that feedback improved performance on average, yet made it worse in 38% of cases. What separated the wins from the losses: feedback aimed at the task helps, feedback aimed at the person hurts. A trainer that shows you the solver's action and the exact EV cost of yours, with no judgment attached, is precisely the kind the research says works, and it arrives 282 times an hour.
What Is the 80/20 of Poker Training?
The first few weeks do most of the work. Among players who drilled 3,000 or more decisions, average EV loss fell from 0.241 BB to 0.170 BB, and 81% of that entire drop had already happened by hand 750. The gains front-load: the first 750 focused reps deliver most of the drop, and the rest is refinement.
How Fast Do Ratings Climb?
Fast: the average rating nearly doubles inside 1,000 decisions. The in-app rating is an Elo-style score that rises with correct decisions and falls with mistakes, weighted by difficulty. Across the same 99-player cohort it climbed from 1,243 to 2,236, rising in every single 100-decision block. It is a progress metric rather than a bankroll metric, but it means the improvement is steady, not a lucky streak.
Is Poker Even a Skill Game?
It is a skill game once the sample gets real, and that has been measured. Researchers analyzed 456 million player-hand observations from real-money online games and found that winning players keep winning: top performers reappear at the top, period after period. Their simulations showed that once performance is measured over roughly 1,500 hands, skill dominates chance (the exact critical point: 1,471 hands). The same dataset shows what most players never fix: only 32% of all players finished with any profit after rake, and rake alone flipped 5.5% of the pool from winners to losers. Luck decides your night. Skill decides your year.
Live tournament data agrees. University of Chicago economists Levitt and Miles tracked all 32,496 entrants of the 2010 World Series of Poker. The 720 players identified beforehand as high-skill earned a +30.5% average return on investment; everyone else averaged -15.6%. That is a 46-point gap between players who study and players who just play, in the most luck-heavy format poker has.
Why Learn GTO Specifically?
Because when game-theory optimal strategy met the best humans on earth, it was not close. DeepStack beat professionals by a variance-adjusted 486 milli-big-blinds per hand (Science, 2017). Libratus beat four elite heads-up specialists by 147 mbb per hand over 120,000 hands at 99.98% statistical significance (Science, 2018). Pluribus then won six-handed against elite pros at roughly 5 BB per 100 (Science, 2019). For scale, professionals treat 50 mbb per hand as a decisive edge. If you are new to the concept, start with what GTO poker actually means.
Why Does Drilling Beat Watching Videos?
Because your brain keeps what it retrieves, not what it re-reads. In the landmark test-enhanced learning study, students who were repeatedly tested recalled 61% of the material a week later; students who repeatedly re-studied recalled 40%, despite reading it four times more often, and despite being more confident they would remember it. Passive review only feels productive.
Poker also happens to sit in the single most trainable category of skill ever measured. A meta-analysis across five domains found deliberate practice explains 26% of the performance differences between players in games, more than in music (21%), sports (18%), education (4%) or professions (under 1%).
Structured drill apps compress this loop harder than any classroom. An efficacy study commissioned by Duolingo found the average learner needed 34 hours to cover a full US college semester of Spanish (the median learner needed more, but the direction is not in dispute). Same mechanism, different game: short reps, instant grading, adaptive difficulty. For a concrete weekly plan built on these principles, see how to practice poker.
What Should You Do With This?
The data suggests a routine that is almost embarrassingly light:
- Train most days, briefly. Ten or more days a month separated the 6x improvers from everyone else. The median session in our data is under 5 minutes.
- Chase the big leaks first. One 2.75 BB blunder outweighs dozens of tiny imprecisions, and big-mistake EV is exactly what fell 38% in the first 1,000 hands.
- Get through your first 750 reps. That is where 81% of the measured improvement lives.
- Drill, do not just watch. Retrieval with instant feedback is the highest-leverage learning mechanism known, and it is the entire design of the GTO Gecko trainer.
Frequently Asked Questions
How many hands of training does it take to improve at poker?
Less than most players think. In GTO Gecko data, mistake rate is measurably lower within 25 short sessions, and 81% of the EV-loss improvement measured across 3,000 hands had already arrived by hand 750. Published research puts the point where skill dominates luck at roughly 1,500 played hands.
Do poker trainers actually work?
The measurable answer from 673,118 graded decisions: players cut their mistake rate 27% by session 25, cut EV loss per decision 28% within 1,000 hands, and 1 in 3 dedicated players cut EV loss in half. Learning science explains the mechanism: retrieval practice with instant feedback beats passive study by a wide, replicated margin.
How long should a poker study session be?
Short and frequent wins. The median training session in our data lasts 4.4 minutes, and players who trained on 10+ days in their first month improved 6 times faster than players who crammed on 4 days or fewer. This matches the spacing literature: distributed practice reliably beats massed practice.
Is poker more luck or skill?
Over a session, luck. Over a sample, skill. The PLOS ONE study of 456 million hands found skill dominates chance once performance is measured over about 1,500 hands, and WSOP data shows high-skill players earning +30.5% ROI while everyone else lost money.